, Brownian motion).
Action learning provides a reusable path objective from trajectory data.
The LEAP framework recovers stochastic actions by comparing observed and reference trajectories, enabling downstream inference and control without retraining.
Big Tech
Sarah Liaw · Yuanqi Du · Yilun Du
Kempner Institute · Harvard University · Microsoft Research New England
Research Digest··2 min read
The authors formalize action learning as recovering a stochastic action from trajectory observations.
§